CN102267095A - Method for monitoring and dressing grinding wheel on line - Google Patents

Abstract

The invention discloses a method for monitoring and dressing a grinding wheel on line. The method comprises the following steps of: detecting acoustic emission (AE) digital signals which are sent by the grinding wheel in a grinding process and processed by using the conventional grinding wheel dressing system; performing fast Fourier transformation on N AE digital signals a(n) in the previous step to obtain different spectral components of N/2 real parts AR(k) and imaginary parts AI(k); and calculating by formulae (1, 2 and 3) to obtain a root-mean-square value (Y), total energy (E), and a center frequency (fc) of a frequency domain, which serve as input samples of a back propagation (BP) neural network, of the AE digital signals. In the method, the root-mean-square value (Y), the total energy (E) and the center frequency (fc) of the frequency domain of the AE digital signals serve as feature extraction values for the first time and then serve as input values of the BP neural network; and the starting time and ending time for dressing of the grinding wheel are accurately predicted by training and test of the BP neural network, so an automatic dressing function of the grinding wheel is realized. The method is high in prediction accuracy.

Description

A kind of emery wheel on-line monitoring and method for trimming

Technical field

The invention belongs to the grinding machine processing control method, be specifically related to a kind of emery wheel on-line monitoring and method for trimming.

Background technology

The finishing of emery wheel has great influence to the grinding performance and the ground effect of emery wheel.Along with grinding develops to intelligent, Self Adaptive Control, requirement can automatically detect the passivation of emery wheel and the accurate moment that crushing is finished, and can in time begin and finish the finishing of emery wheel, to improve grinding quality and production efficiency.In work in the past, for fear of the workpiece grinding burn, adopt regularly finishing mostly, when emery wheel does not reach the working life limit as yet, just in advance it is repaired, be blindness (" grinding quality Study on on-line monitor method ", Liu Guijie etc. like this, diamond and grinding materials and grinding tool engineering, 2004.10).Frequent trimming wheel not only can influence working (machining) efficiency, also can accelerate the loss of emery wheel, particularly uses cubic boron nitride (CBN) emery wheel costliness, has increased production cost.Otherwise, if incured loss through delay the finishing cycle, then can influence the precision and the surface quality of workpiece again, cause waste product.Therefore, can be accurately and timely trimming wheel automatically, be to improve grinding productivity ratio, guarantee an important channel of Grinding Machining Quality.

In developmental research in recent years, acoustic emission (AE) signal is widely adopted as the information source of grinding control.Many results of study show: the amplitude by monitoring AE signal changes, and can estimate the sharpness of emery wheel and the working life of definite emery wheel (" based on the monitoring of line again of the abrasive grinding wheel state of neutral net ", Liu Guijie etc., Northeastern University's journal, 2002.10; " application of STFT in the AE signal characteristic extracts ", Liao Chuanjun etc., Chinese journal of scientific instrument, 2008.9; " AE of grinding process monitors model " Wu Guoyue, grinding machine and grinding, 1999.1).But, when processing work material, processing conditions and machined parameters often change,, only depend on the amplitude size that detects acoustic emission signal can't judge the emery wheel degree of passivation because the amplitude of acoustic emission signal also can acute variation take place thereupon.For this reason, this paper has proposed the method for FFT spectrum analysis, and promptly the high frequency range component that sharp time processing produces according to emery wheel is big and spectrum component is many, and high frequency range component is little few with spectrum component during passivation.We set a band limit, when radio-frequency component is less than certain value, just need trimming wheel, when high fdrequency component reaches certain scope, finish the finishing of emery wheel, realize the automatic dressing of emery wheel.

Summary of the invention

The purpose of this invention is to provide a kind of emery wheel on-line monitoring and method for trimming.

Emery wheel on-line monitoring and method for trimming that the present invention proposes, realize by following steps:

Step 1, utilize existing crushing system to detect that emery wheel sends and treated acoustic emission (AE) data signal in grinding process;

Step 2, with N acoustic emission (AE) the data signal a (n) of step 1, obtain by FFT FFT

The different spectral component of individual real part AR (k) and imaginary part AI (k);

Step 3, pass through formula (1), (2), (3) again and calculate, obtain root-mean-square value Y, gross energy E, the frequency domain centre frequency fc of acoustic emission (AE) data signal and as the input sample of BP neutral net:

$Y=\sqrt{\frac{\underset{n=1}{\overset{N}{\mathrm{\Σ}}}{a}^2\left(n\right)}{N}}---\left(1\right)$

$E=\underset{k=0}{\overset{\frac{N}{2}}{\mathrm{\Σ}}}(\mathrm{AR}{\left(k\right)}^2+\mathrm{AI}{\left(k\right)}^2)---\left(2\right)$

$f_c=\frac{\underset{k=0}{\overset{\frac{N}{2}}{\mathrm{\Σ}}}k\×f\×(\mathrm{AR}{\left(k\right)}^2+\mathrm{AI}{\left(k\right)}^2)}{2\mathrm{\πE}}---\left(3\right)$

In the formula: Y is the root-mean-square value of acoustic emission (AE) data signal, a (n) is acoustic emission (AE) data signal, N is input acoustic emission (AE) digital signal samples number, k is 0,1 ..., N, AR (k) calculates back real part frequency spectrum by FFT, AI (k) calculates back imaginary part frequency spectrum by FFT, and E is the gross energy of acoustic emission (AE) data signal, f _cThe frequency domain centre frequency of acoustic emission (AE) data signal, f is the resolution frequency of EFT spectrum analysis, is f as acoustic emission (AE) signal highest frequency _Max, sample frequency is f _s, sampled point is chosen N, and then differentiating frequency is f=f _s/ N.

Step 4, BP neutral net are got a part with the sample of step 3 and are learnt and train, and make error amount that sample training produces within the scope of setting, and obtain the weights V of described BP neutral net _Ij, W _JkWith controlling value R _j, Q _kAs the standard value of BP neutral net matrix, again samples remaining is inputed to described BP neutral net matrix and test the numerical value that draws emery wheel passivation and the following two kinds of different range of sharp state, respectively in order to predict the initial moment and the finish time of crushing;

Step 5, utilize VB, Mitrix-VB Software tool that step 2,3,4 process are programmed and calculate the centre frequency f of root-mean-square value Y, signal energy E, signal _cWeights V with the BP neutral net _Ij, W _Jk, controlling value R _j, Q _kWith the numerical value of emery wheel passivation and the following two kinds of different range of sharp state, the re-set target so that whether the result of verification step 4 reaches the emery wheel automatic dressing in the camshaft grinding machine digital control system, realizes emery wheel automatic dressing function with program portable.

The inventive method is first with root-mean-square value (Y), gross energy (E), the frequency domain centre frequency (f of acoustic emission (AE) data signal _c) as the feature extraction value, then with its input value as the BP neutral net, this method prediction accuracy height has improved machining accuracy and working (machining) efficiency, has also prolonged emery wheel service life, has reduced enterprise's production cost.

The invention will be further described below in conjunction with the drawings and specific embodiments.

Description of drawings

Fig. 1 is a camshaft grinding machine emery wheel automatic dressing system block diagram.

Fig. 2 is BP neural network model figure.

Fig. 3 is a BP neural metwork training program flow diagram.

Fig. 4 is a BP neutral net test program flow chart.

Fig. 5 is BP neural network parameter input figure.

Fig. 6 is BP neural metwork training operation result figure.

Fig. 7 is BP neural metwork training operation result figure.

The specific embodiment

Be illustrated in figure 1 as existing YTMCNC8326 numerically controlled cam shaft grinder wheel conditioning system block diagram, it is made up of parts such as emery wheel, AE sensor, emery wheel trimmer, work holder, MARPOSS P7WB dynamic balance instrument, Siemens 840D digital control system, Siemens 611D drivers.The AE sensor is installed in the dynamic/static piezoelectric main shaft axle center, the MARPOSS dynamic balance instrument detects the AE signal, after it is carried out digitlization and filtering and handles, send NCU by the Profibus bus, send MMC user interface (OEM) through NCU, perhaps send MMC user interface (OEM) by serial ports COM.In OEM, carry out spectrum analysis, according to the frequency spectrum difference of different phase, the moment of control beginning or end crushing, the automatic dressing of realization emery wheel.

The step and the principle of the inventive method are as follows:

Step 1, utilize existing crushing system to detect that emery wheel sends and treated acoustic emission (AE) data signal in grinding process;

Step 2, with N acoustic emission (AE) the data signal a (n) of step 1, obtain by FFT FFT

The different spectral component of individual real part AR (k) and imaginary part AI (k);

The fft algorithm principle

In digital signal processing, DFT (Discrete Fourier Transform is called for short DFT) provides a kind of mathematical method of utilizing digital computer to make Fourier transformation computation.The DFT calculating formula is:

$X\left(k\right)=\mathrm{DFT}\left[x\left(n\right)\right]=\underset{n=0}{\overset{N-1}{\mathrm{\Σ}}}x\left(n\right)W_N^{\mathrm{nk}},0\≤k\≤N-1---\left(4\right)$

Wherein,

Be called the Fourier factor.

But the amount of calculation of DFT is very big, and FFT (Fast Fourier Transform, be called for short FFT) algorithm is the algorithm of a kind of DFT of minimizing computing time of developing on the DFT basis, and it has improved operation efficiency widely.The present invention is directed to the canonical form (being Cooley-Tukey algorithm) of FFT, it is basic 2FFT algorithm, and it is 2 M power that this algorithm is got N: N=2 ^M, M is a positive integer, the starting point of algorithm is that N some DFT computing is decomposed into two groups

The DFT computing of individual point, calculating formula is:

$X\left(k\right)=\underset{r=0}{\overset{\frac{N}{2}-1}{\mathrm{\&Sigma;}}}x\left(2r\right)W_{\mathrm{<figure-callout\; id="2"\; label="N"\; filenames="HDA0000086182570000032.png,HDA0000086182570000041.png"\; state="\{\{state\}\}">N</figure-callout>}}^{2\mathrm{rk}}+\underset{r=0}{\overset{\frac{N}{2}-1}{\mathrm{\&Sigma;}}}x(2r+1)W_N^{(2r+1)k}---\left(5\right)$

Wherein, the W implication is the same; R=0,1 ...,

2r represents even number, and 2r+1 represents odd number.

By said method, in even number, odd number, can be divided into four groups of even number, odd numbers again, carry out above-mentioned computing, the rest may be inferred.

Can obtain the spectrum distribution of various signals by above-mentioned algorithm.

Step 3, the well-known feature space of use are explained method as suitable " feature ", the root-mean-square value Y of acoustic emission (AE) data signal, frequency domain centre frequency f _cGo out with gross energy E is selected, calculate, obtain root-mean-square value Y, gross energy E, the frequency domain centre frequency f of acoustic emission (AE) data signal through formula (1), (2), (3) _cAnd as the input sample of BP neutral net:

$Y=\sqrt{\frac{\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}{a}^2\left(n\right)}{N}}---\left(1\right)$

$E=\underset{k=0}{\overset{\frac{N}{2}}{\mathrm{\&Sigma;}}}(\mathrm{AR}{\left(k\right)}^2+\mathrm{AI}{\left(k\right)}^2)---\left(2\right)$

$f_c=\frac{\underset{k=0}{\overset{\frac{N}{2}}{\mathrm{\&Sigma;}}}k\&times;f\&times;(\mathrm{AR}{\left(k\right)}^2+\mathrm{AI}{\left(k\right)}^2)}{2\mathrm{\&pi;E}}---\left(3\right)$

In the formula: Y is the root-mean-square value of acoustic emission (AE) data signal, a (n) is acoustic emission (AE) data signal, N is input acoustic emission (AE) digital signal samples number, k is 0,1 ..., N, AR (k) calculates back real part frequency spectrum by EET, AI (k) calculates back imaginary part frequency spectrum by FFT, and E is the gross energy of acoustic emission (AE) data signal, f _cThe frequency domain centre frequency of acoustic emission (AE) data signal, f is the resolution frequency of FFT spectrum analysis, is f as acoustic emission (AE) signal highest frequency _Max, sample frequency is f _s, sampled point is chosen N, and then differentiating frequency is f=f _s/ N.

Step 4, described BP neutral net are got a part with the sample of step 3 and are learnt and train, and make error amount that sample training produces within the scope of setting, and obtain the weights V of described BP neutral net _Ij, W _JkWith controlling value R _j, Q _kAs the standard value of BP neutral net matrix, again samples remaining is inputed to described BP neutral net matrix and test the numerical value that draws emery wheel passivation and the following two kinds of different range of sharp state, respectively in order to predict the initial moment and the finish time of crushing;

The foundation of BP neural network model:

(1) topological structure of neutral net

Neutral net can be the Complex Nonlinear System opening relationships model of many inputs and many outputs, is highly suitable for replacing human decision behavior based on experience.

The application adopts three layers of BP neutral net as emery wheel state recognition model, is illustrated in figure 2 as general three layers of BP neural network model figure.In fact the selection of hidden layer node number and random following, solution is to increase the neuron number from small to large gradually, until obtaining higher precision and convergence rate faster.This paper selects self-structuring type neutral net for use, gives hidden layer node of network earlier, calculates its error; And then add a hidden layer node; If satisfy, then withdraw from, the hidden layer node number of network is determined.Carry out training result according to this method and show that activation primitive is the Sigmoid function, topological structure is the neutral net of 3-15-1, and recognition accuracy is about 98%, so the application has selected this topological structure for use.

(2) Learning Algorithm

Suppose given pattern x _iBe network input, O _kBe network output, d _kBe desired value, output layer is all had j=0,1 ..., m; K=1,2 ... l; Hidden layer all there is i=0,1 ..., n; J=1,2 ..., m.

$E_p=\frac{1}{2}\mathrm{\&Sigma;}{(d_k-O_k)}^2---\left(6\right)$

d _kBe desired value, O _kBe output valve, E _pBe the error amount quadratic sum

For output layer,

Deployablely be:

${\mathrm{\&delta;}}_k^v=(d_k-O_k)f^{\&prime;}\left({\mathrm{net}}_k\right)---\left(7\right)$

Wherein be $f\left(x\right)=\frac{1}{1+e^{-x}},$

Output error signal.

For hidden layer,

Deployablely be:

${\mathrm{\&delta;}}_j^y=f^{\&prime;}\left({\mathrm{net}}_j\right)\underset{k=1}{\overset{l}{\mathrm{\&Sigma;}}}{\mathrm{\&delta;}}_k^v{w}_{\mathrm{jk}}---\left(8\right)$

Wherein

The hidden layer error signal, W _JkThe output layer weights

$f^{\&prime;}\left({\mathrm{net}}_k\right)=\frac{{\&PartialD;O}_k}{\&PartialD;{\mathrm{net}}_k}=O_k\left({1-O}_k\right)$

$f^{\&prime;}\left({\mathrm{net}}_j\right)=\frac{{\&PartialD;y}_j}{\&PartialD;{\mathrm{net}}_j}=y_j\left({1-y}_j\right)$

So have:

${\mathrm{\&delta;}}_k^v=(d_k-O_k)O_k\left({1-O}_k\right)---\left(9\right)$

${\mathrm{\&delta;}}_j^y=y_j(1-y_j)\underset{k=1}{\overset{l}{\mathrm{\&Sigma;}}}{\mathrm{\&delta;}}_k^v{w}_{\mathrm{jk}}---\left(110\right)$

Wherein: yj is the hidden layer output valve.

Output layer weights correction is

Aw _jk＝η(d _k-O _k)O _k(1-O _k)y _j (11)

Hidden layer weights correction is

$\mathrm{\&Delta;}{v}_{\mathrm{ij}}=\mathrm{\&eta;}\left(\underset{k=1}{\overset{l}{\mathrm{\&Sigma;}}}{\mathrm{\&delta;}}_k^v{w}_{\mathrm{jk}}\right)y_j\left({1-y}_j\right)x_i---\left(12\right)$

In order to improve the network convergence characteristic, select the innovation representation of Δ W (n) for use at this

ΔW(n)＝ηδX+αW(n-1)

Wherein, W is certain a layer of weight matrix, and X is certain a layer of input vector, and η is the study factor; α is for adjusting the proportionality constant of variable quantity.

(3) the VB6.0 programming realizes the training and the test of FFT spectrum analysis and BP neutral net

Crushing control interface is mainly used in gathers the AE signal, by the AE signal is carried out spectrum analysis, AE signal spectrum figure draws, extract the feature root-mean-square value Y of AE signal, the centre frequency fc of frequency domain amplitude, gross energy E, by the BP neural network prediction, judge the starting and ending moment of crushing.

Receive digital AE signal from dynamic balance instrument by serial ports or Profibus bus, numeral AE signal is divided into 4 groups: the 1st group of digital AE signal of gathering when being the sharp grinding work piece of emery wheel, the 2nd group of digital AE signal of gathering when being emery wheel passivation grinding work piece, the 3rd group of digital AE signal of gathering when being emery wheel finishing passivation emery wheel, the 4th group is the digital AE signal that emery wheel is gathered when repairing sharp emery wheel.Each group is gathered 200 sections, gathers 256 sample points for every section.Respectively above-mentioned 4 groups of AE signals are carried out spectrum analysis, calculate AE signal root-mean-square value Y, the centre frequency fc of frequency domain amplitude, gross energy E characteristic value, then the 1st, 2 stack features values are trained neutral net as BP neutral net input value, in error amount reaches the scope of allowing, can utilize this Network Recognition wheel grinding workpiece to obtain the initial moment that emery wheel need be repaired.Adopt the characteristic value that uses the same method to the 3rd, 4 group and train, can obtain crushing finish time.

Be illustrated in figure 3 as BP neural metwork training program flow diagram, be illustrated in figure 4 as BP neutral net test program flow chart.

Step 5, utilize VB, Mitrix-VB Software tool that step 2,3,4 process are programmed and calculate the centre frequency f of root-mean-square value Y, signal energy E, signal _cWeights V with the BP neutral net _Ij, W _Jk, controlling value R _j, Q _kWith the numerical value of emery wheel passivation and the following two kinds of different range of sharp state, the re-set target so that whether the result of verification step 4 reaches the emery wheel automatic dressing in the camshaft grinding machine digital control system, realizes emery wheel automatic dressing function with program portable.

Application example

(1) reads the AE data signal from the MARPOSS dynamic balance instrument by serial port, digital AE signal is divided into 4 groups: the 1st group of digital AE signal of gathering when being the sharp grinding work piece of emery wheel, the 2nd group of digital AE signal of gathering when being emery wheel passivation grinding work piece, the 3rd group of digital AE signal of gathering when being emery wheel finishing passivation emery wheel, the 4th group is the digital AE signal that emery wheel is gathered when repairing sharp emery wheel.Each group is gathered 200 sections, gathers 256 sample points for every section.

(2) with 4 groups, every group 200 sections, every section 256 sampled points by the formula (5) in the step 2, carry out FFT FFT, by step 5 pair formula (5) programming, can calculate 4 groups, every group 200 sections, every end respectively by the different spectral component of real part AR (0)～AR (127), imaginary part AI (0)～AI (127).

(3) with 4 groups, every group 200 sections, every section 256 sampled points, 4 groups, every group 200 sections, every end are respectively by the different spectral component of real part AR (0)～AR (127), imaginary part AI (0)～AI (127).Calculate by the formula in the step 3 (1), (2), (3) respectively, by step 5 pair formula (1), (2), (3) programming, can calculate root-mean-square value Y, frequency domain centre frequency fc, the gross energy E characteristic value of 4 groups, every group 200 acoustic emissions (AE) data signal, wherein be one group part calculated value wherein shown in the table 1.f _s=250kHz, N=1024, then: f=f _s/ N=250000/1024=244Hz.

The characteristic value that table 1 extracts

(4) respectively above-mentioned 4 stack features values being divided into two groups carries out the BP neutral net and trains, the root-mean-square value Y of one group of 200 acoustic emissions (AE) data signal during by the 1st group the sharp grinding work piece of emery wheel, centre frequency fc, the root-mean-square value Y of 200 acoustic emissions (AE) data signal during the emery wheel passivation grinding work piece of gross energy E characteristic value and the 2nd group, frequency domain centre frequency fc, gross energy E characteristic value is formed one group, 200 root-mean-square value Y that choose 160 acoustic emissions (AE) data signal when in like manner, another group is repaired the passivation emery wheel by the 3rd group emery wheel, frequency domain centre frequency fc, gross energy E characteristic value and the 4th group are emery wheel 200 root-mean-square value Y that choose 160 acoustic emissions (AE) data signal when repairing sharp emery wheel, centre frequency fc, gross energy E characteristic value is formed one group.The formula of sending in the step (4) calculates, by formula programming in the step 5 pair step (4), the root-mean-square value Y of 160 AE signals during with the 1st group in one group the sharp grinding work piece of emery wheel, frequency domain centre frequency fc, the root-mean-square value Y of 160 AE signals during the emery wheel passivation grinding work piece of gross energy E characteristic value characteristic value and the 2nd group, frequency domain centre frequency fc, gross energy E characteristic value is sent BP neutral net, train, ideal output when emery wheel is sharp is set to 1, ideal output during the emery wheel passivation is set to 0, then train, in the scope that error curve is reached allow.Train at neutral net interface with the VB6.0 programming, is illustrated in figure 5 as BP neural network parameter input figure, is illustrated in figure 6 as BP neural network computing figure as a result.Annotate: AE signal root-mean-square value Y, AE signal frequency domain center frequency value fc * 0.001, AE signal energy value E * 0.1.

(5) weights and the threshold values that come out of storage neural metwork training (shown the V among the BP neural metwork training operation result figure as shown in Figure 6 _Ji, W _Kj, R _j, Q _k), take out remaining 40 groups then neutral net tested, judge emery wheel when processing work be in sharp, the still correctness in the passivation moment, operation result is illustrated in figure 7 as BP neutral net test result figure.In like manner, root-mean-square value Y, the frequency domain centre frequency fc of 160 AE signals when using another group by the 3rd group emery wheel finishing passivation emery wheel, gross energy E characteristic value characteristic value and the 4th group are that root-mean-square value Y, frequency domain centre frequency fc, the gross energy E characteristic value of emery wheel 200 160 acoustic emissions (AE) data signal when repairing sharp emery wheel carried out training and testing to another neutral net network, judge that emery wheel is to be in passivation when finishing, still the correctness in the sharp moment.To realize the automatic dressing function of emery wheel.

The inventive method adopts the acoustic emission AE data signal of dynamic balance instrument output alone, it is carried out the FFT spectrum analysis, calculate root-mean-square value Y, frequency domain centre frequency fc, the gross energy E characteristic value of acoustic emission (AE) data signal, send the BP neutral net to train and test then, obtain the sharp accurate moment of emery wheel passivation accurately and crushing, realize the online grooming function of emery wheel.With former method ratio, have intellectuality, emery wheel prediction accuracy height has improved machining accuracy and working (machining) efficiency, has prolonged emery wheel service life, has reduced enterprise's production cost.

Claims (1)

1. emery wheel on-line monitoring and method for trimming is characterized in that being realized by following steps:

Step 1, utilize existing crushing system to detect that emery wheel sends and treated acoustic emission (AE) data signal in grinding process;

Step 2, with N acoustic emission (AE) the data signal a (n) of step 1, obtain by FFT FFT The different spectral component of individual real part AR (k) and imaginary part AI (k);

Step 3, pass through formula (1), (2), (3) again and calculate, obtain root-mean-square value Y, gross energy E, the frequency domain centre frequency f of acoustic emission (AE) data signal _cAnd as the input sample of BP neutral net:

$Y=\sqrt{\frac{\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}{a}^2\left(n\right)}{N}}---\left(1\right)$

$E=\underset{k=0}{\overset{\frac{N}{2}}{\mathrm{\&Sigma;}}}(\mathrm{AR}{\left(k\right)}^2+\mathrm{AI}{\left(k\right)}^2)---\left(2\right)$

$f_c=\frac{\underset{k=0}{\overset{\frac{N}{2}}{\mathrm{\&Sigma;}}}k\&times;f\&times;(\mathrm{AR}{\left(k\right)}^2+\mathrm{AI}{\left(k\right)}^2)}{2\mathrm{\&pi;E}}---\left(3\right)$

In the formula: Y is the root-mean-square value of acoustic emission (AE) data signal, a (n) is acoustic emission (AE) data signal, N is input acoustic emission (AE) digital signal samples number, k is 0,1 ..., N, AR (k) calculates back real part frequency spectrum by FFT, AI (k) calculates back imaginary part frequency spectrum by FFT, and E is the gross energy of acoustic emission (AE) data signal, f _cThe frequency domain centre frequency of acoustic emission (AE) data signal, f is the resolution frequency of FFT spectrum analysis, is f as acoustic emission (AE) signal highest frequency _Max, sample frequency is f _s, sampled point is chosen N, and then differentiating frequency is f=f _s/ N.

Step 4, BP neutral net are got a part with the sample of step 3 and are learnt and train, and make error amount that sample training produces within the scope of setting, and obtain the weights V of described BP neutral net _Ij, W _JkWith controlling value R _j, Q _kAs the standard value of BP neutral net matrix, again samples remaining is inputed to described BP neutral net matrix and test the numerical value that draws emery wheel passivation and the following two kinds of different range of sharp state, respectively in order to predict the initial moment and the finish time of crushing;

Step 5, utilize VB, Mitrix-VB Software tool that step 2,3,4 process are programmed and calculate the centre frequency f of root-mean-square value Y, signal energy E, signal _cWeights V with the BP neutral net _Ij, W _Jk, controlling value R _j, Q _kWith the numerical value of emery wheel passivation and the following two kinds of different range of sharp state, the re-set target so that whether the result of verification step 4 reaches the emery wheel automatic dressing in the camshaft grinding machine digital control system, realizes emery wheel automatic dressing function with program portable.

Images